Quantum compiling by deep reinforcement learning

Quantum compiling, approximating any unitary transformation as a sequence of gates from a finite universal set, faces a trade-off between sequence length, precompilation time, and execution time. Traditional methods are slow, unsuitable for real-time computation. This paper proposes a deep reinforcement learning (DRL) method requiring a single precompilation to learn a strategy for approximating single-qubit unitaries. The approach reduces execution time, improves the length-execution time trade-off, and enables potential real-time operations.

Lorenzo Moro, Matteo G. A. Paris, Marcello Restelli, Enrico Prati